[629] in Info-AFS_Redistribution
Re: tar vs. cp in AFS
daemon@ATHENA.MIT.EDU (Phil_Hirsch@transarc.com)
Mon Feb 24 13:44:05 1992
Date: Mon, 24 Feb 1992 12:15:16 -0500 (EST)
From: Phil_Hirsch@transarc.com
To: Info-AFS@transarc.com, sandman@bu-it.bu.edu (Dominick M. Galang)
In-Reply-To: <9202241625.AA14166@buit52.bu.edu>
This sounds like an AFS bug that has surfaced here before. The bug is
this: if you open a file with certain combinations of flags (O_WRONLY
being the most important, as I recall), and if the file's Unix mode
bits allow writing but not reading, then the open will fail. Apparently
some implementations of tar are affected by this.
This topic was discussed in early February 1991 on this mailing list;
Look for a thread titled "A question on protection semantics." I can
send you a summary if you need it. I posted this description of the bug
in that discussion:
I think the tar problem you're seeing is a manifestation of a known bug in
AFS. It shows up under SunOS, HP/UX, and AOS. If you attempt to call open()
on an existing file whose Unix mode bits allow writing but not reading, and
if you use open() flags O_WRONLY|O_CREAT, the open() fails with errno 13
(EACCES). On Unix filesystems, the open() succeeds, as it should.
I don't know specifically how tar opens files, but your symptoms seem to
match those that tickle this bug.
- Phil H.
